01 / Read · Evidence selected
Start from the artifact trail.
Inspect the workflow, method, and implementation receipts behind each claim.
Evidence to inspect
Start from operator friction, runtime behavior, and implementation receipts. Papers, experiments, and field notes stay tied to the workflow and make the claim, artifact, methodology, and next move inspectable.
- Paper or experiment names the source workflow and operating question
- Claim is tied to methodology, implementation notes, or the research graph
- Operator notes explain why the workflow exists and where it breaks
paper archive / methodology / research graph
02 / Validate · Runtime proof needed
Move the claim into a live surface.
Test timing, state, and failure behavior in the workbench.
Evidence to inspect
Compare cost, speed, and maintenance drag across AI-native stacks. When the claim still depends on timing, state, or failure behavior, move it into the workbench instead of adding another paragraph.
- Pattern has a concrete execution question
- Timing, state, and failure behavior are visible
- Implementation tradeoffs are written for operators, not leaderboard chatter
runtime note / motion output / data trace
03 / Scope · Delivery decision
Carry proven evidence into delivery.
Give the delivery handoff explicit controls, recovery, and an owner.
Evidence to inspect
Turn proven judgment into policy packs, release checks, contracts, and runbooks. When the risk is commercial, operational, or reputational, route the evidence to a scoped workflow with an explicit owner.
- Database / Automation / Judgment remains the operating frame
- Evidence points to controls, policy, and recovery
- Delivery handoff has an owner and a clear first lane
handoff note / policy cue / mapping session